{"id":"3d2624ae-ac0c-4986-b0e2-c88167e0d62d","arxiv_id":"2511.09102","paper_version":3,"verdict":"REJECT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"Coherent measurements are claimed to be exactly those that can demonstrate SDI steering, enabling randomness certification from separable isotropic states even at arbitrarily low detection efficiency.","lead":"The paper claims a theorem: a set of quantum measurements can demonstrate semi-device-independent (SDI) steering if and only if the measurements are coherent (noncommuting), and uses this to certify local randomness from states that are not even entangled. The result matters for quantum random number generation because it promises a way around entanglement certification and low detection efficiency—but the central proof rests on a step that is not valid.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Converse of Theorem 1 fails: Eq. (A6)–(A7) does not force hidden states to be eigenprojectors of ρ_B; an explicit qutrit LHS model has noncommuting SEO, refuting Cor. 2.","rationale":"The paper's headline result is the iff between coherence of a measurement assemblage and its ability to demonstrate SDI steering (Cor. 1) and the equivalent statement in terms of noncommuting SEO (Cor. 2). The only nontrivial direction is the converse of Theorem 1: from a dimensionally-restricted LHS model to commuting/incoherent SEO. That step is exactly Eq. (A6)-(A7), where the authors replace the true condition Σ_λ τ_λ=1 with the much stronger claim that each τ_λ is a projector onto a vector of a basis diagonalizing ρ_B. No such conclusion follows. The qutrit example above satisfies the Definition 1 LHS condition and has B_{0|0}=σ_0 and B_{0|1}=σ_1 with nonzero commutator. Thus the 'if' clause of Corollary 2 is violated. Since Corollary 2 is used to justify faithfulness of S_Υ and the monotonicity argument, and since the randomness bound (21) is derived from S_Υ, the downstream quantitative claims inherit the failure. I do not see a way to repair the main theorem without either changing Definition 1 or adding strong restrictions (e.g., requiring hidden states to be mutually orthogonal pure states), which is not the stated framework. Therefore the reader's REJECT verdict is correct; no adjustment.","tokens_in":13934,"tokens_out":9970,"duration_ms":87118,"concrete_test":"Instantiate the qutrit example: take ρ_B=I, σ_0=diag(0.3,0.1,0), σ_1=0.4|+⟩⟨+|+0.2|2⟩⟨2| with |+⟩=(|0⟩+|1⟩)/√2, σ_2=I−σ_0−σ_1. Verify σ_0,σ_1,σ_2≥0 and that the assemblage σ_{0|0}=σ_0, σ_{1|0}=σ_1+σ_2, σ_{0|1}=σ_1, σ_{1|1}=σ_0+σ_2 admits the LHS decomposition of Definition 1 with d_λ=3. Then compute the SEOs B_{0|0}=σ_0, B_{0|1}=σ_1 and check [σ_0,σ_1]≠0. This directly falsifies the converse of Theorem 1 and Corollary 2.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of the converse of Theorem 1 (Appendix A, Eqs. A5–A9) assumes that any LHS decomposition σ_{a|x}=Σ_λ p(a|x,λ)σ_λ with d_λ≤d_A, Σ_λ σ_λ=ρ_B, must have σ_λ=ρ_B^{1/2}|λ⟩⟨λ|ρ_B^{1/2} for a common eigenbasis of ρ_B. This inference is invalid: Eq. (A6) only says τ_λ:=ρ_B^{-1/2}σ_λρ_B^{-1/2} are positive operators summing to 1; they need not be rank-one projectors onto a common orthonormal basis, and positive decompositions of the identity can be noncommuting. Explicit counterexample with d_A=3 and ρ_B=1: take σ_0=diag(0.3,0.1,0), σ_1=0.4|+⟩⟨+|+0.2|2⟩⟨2| with |+⟩=(|0⟩+|1⟩)/√2, σ_2=1−σ_0−σ_1. Then σ_j≥0 and Σ_j σ_j=1. Define x=0,1 outcomes {0,1}: σ_{0|0}=σ_0, σ_{1|0}=σ_1+σ_2, σ_{0|1}=σ_1, σ_{1|1}=σ_0+σ_2. This is a valid LHS model with d_λ=3≤d_A, so no SDI steering. But the SEO are B_{0|0}=σ_0, B_{0|1}=σ_1, and [σ_0,σ_1]≠0; hence Corollary 2's 'only if' direction fails. The central equivalence, resource monotone, and randomness claims all rest on this step, so the central claim is unsupported and in fact false as stated.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper claims to give a complete operational characterization of coherent (pairwise noncommuting) measurement assemblages via semi-device-independent (SDI) steering. The central results are: (i) Theorem 1 and Corollary 1 assert that a measurement assemblage can demonstrate SDI steering if and only if it is coherent; (ii) Corollary 2 identifies SDI steering with pairwise noncommutativity of steering-equivalence observables (SEO); (iii) a nonconvex resource theory with monotone S_Υ is constructed from this criterion; (iv) an SDI QRNG is claimed to certify intrinsic randomness from any coherent measurement, with arbitrarily low detection efficiency and without certifying entanglement. The paper also presents explicit calculations for two-qubit isotropic states.","tokens_in":14468,"tokens_out":11085,"duration_ms":113452,"significance":"If the central equivalence were correct, the paper would forge a clean link between coherence of generalized measurements and an operational nonlocality task, extending the known steering-incompatibility correspondence and suggesting practical QRNG advantages at arbitrarily low detection efficiency. The forward direction—incoherent measurements cannot demonstrate SDI steering—is straightforwardly argued via Eqs. (A1)-(A4), and the construction using pure entangled states in Eq. (8) is a useful observation. However, the converse direction, which is the load-bearing step for the equivalence, is invalid. The paper provides no machine-checked proofs or reproducible code, and its central application claims rest on the failed converse. The contribution, as stated, is therefore not established.","major_comments":[{"comment":"The converse proof of Theorem 1 is invalid. From Σ_λ ρ_B^{-1/2}σ_λ ρ_B^{-1/2}=I it does not follow that each ρ_B^{-1/2}σ_λ ρ_B^{-1/2} is a rank-one projector onto a common eigenbasis of ρ_B. Positive decompositions of the identity into d_λ≤d_A terms need not be diagonal in a single basis. Concretely, take d_A=3, ρ_B=I, σ_0=diag(0.3,0.1,0), σ_1=0.4|+><+|+0.2|2><2|, σ_2=I−σ_0−σ_1, and define σ_{0|0}=σ_0, σ_{1|0}=σ_1+σ_2, σ_{0|1}=σ_1, σ_{1|1}=σ_0+σ_2. This is a valid dimensionally-restricted LHS model with d_λ=3≤d_A, yet the SEO satisfy [σ_0,σ_1]≠0. Thus the 'only if' direction of Corollary 2 is false, the converse of Theorem 1 fails, and the faithfulness property of S_Υ (Definition 2, property 1) collapses.","section":"Appendix A, Eqs. (A5)-(A7); Corollary 2"},{"comment":"The randomness claim is refuted by essentially the same construction. The counterexample state assemblage can be produced by a classical-quantum state with a three-dimensional classical register (ρ_AB=Σ_λ |λ><λ|_A ⊗ σ_λ), so a three-dimensional Eve holding λ can predict Alice's outcome exactly. This directly contradicts the assertion that noncommuting SEO certify intrinsic randomness under d_E≤d_A. The proof in Appendix D only argues that commuting SEO are compatible with a CQ state; it does not show that noncommuting SEO exclude a dimension-bounded classical model, and the final paragraph simply stipulates the dimension restriction on Eve without physical or information-theoretic justification. The QRNG application and the claim of 'genuine randomness' are therefore unsupported as stated.","section":"Theorem 3 and Appendix D"},{"comment":"Because Corollary 2 is false, the proposed measure S_Υ in Eq. (11) is not a faithful quantifier of SDI steering: there exist free assemblages (admitting a dimensionally-restricted LHS model) with strictly positive S_Υ. Consequently, the monotonicity result of Theorem 2 and Appendix C, even if formally correct for the noncommutativity measure, does not establish a monotone for the SDI-steering resource. The resource-theoretic interpretation of the paper's results is thus not supported by the provided proof.","section":"Definition 2 and resource-theoretic quantification"}],"minor_comments":[{"comment":"The LHS decomposition in Eq. (A5) writes p(a|x,λ) without the prior p(λ); the normalization is only consistent if σ_λ are unnormalized and absorb p(λ). Please make this explicit to avoid confusion.","section":"Notation, Eq. (A5)"},{"comment":"References [8] and [9] are duplicated; one should be removed.","section":"References"},{"comment":"The derivation of S_Υ = ||r|| ||v|| sin(angle) appears to drop normalization factors from Definition 2. Please check the constants; as written, Eq. (E3) does not obviously follow from Eq. (11).","section":"Appendix E, Eq. (E3)"},{"comment":"The acronym '1SSDI' is nonstandard; a brief definition at first use would improve readability.","section":"General"}],"recommendation":"reject","confidential_remarks":"The counterexample in Appendix A is decisive: it refutes the paper's central equivalence and, via the classical-quantum realization, also the randomness theorem. I do not see a repair within the manuscript's scope; the main claims would need to be substantially restricted (possibly to d_A=2) and the security proof for the QRNG reworked. The authors may still have a valid partial result for qubits, but the present version's central claims are false as stated."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper's main theorem is not just unproven; it is false as stated. The converse direction of Theorem 1 relies on an invalid linear algebra inference in Appendix A: from ∑_λ ρ_B^{-1/2} σ_λ ρ_B^{-1/2} = 1, the authors conclude σ_λ = ρ_B^{1/2}|λ⟩⟨λ|ρ_B^{1/2} for an orthonormal basis diagonalizing ρ_B. That does not follow. Positive decompositions of the identity need not be rank-one eigenprojectors, and the stress-test counterexample is correct. With d_A=3 and ρ_B=I, take σ_0 = diag(0.3,0.1,0), σ_1 = 0.4|+⟩⟨+| + 0.2|2⟩⟨2|, and σ_2 = I − σ_0 − σ_1. All are positive and sum to I. Defining outcomes as in the stress note gives a valid dimensionally-restricted LHS model (d_λ=3≤d_A) whose SEO are σ_0 and σ_1, which do not commute. So Corollary 2 fails, and with it the 'only if' direction of the central equivalence. The faithfulness of the S_Υ monotone rests directly on Corollary 2, so the resource-theoretic quantification is also unsupported. Theorem 3's randomness claim requires a dimension restriction on Eve's purification—a strong assumption that is not highlighted in the abstract's promise of randomness 'beyond entangled states.' The QRNG application is interesting only if the steering criterion holds, and it does not.\n\nWhat the paper does well: it crisply identifies the gap between measurement incompatibility and noncommutativity in the SDI setting, and the construction of a nonconvex monotone from the SEO is a natural extension of the authors' earlier work. The presentation is clear, and the literature engagement is honest. But the load-bearing step is a genuine error, not a minor gap. The counterexample is simple enough to be checked quickly, and it kills the main claims as stated.\n\nI would not cite this paper in its current form. It deserves a serious referee because the question is good and the counterexample is instructive, but the central claims need major revision—likely a corrected theorem that characterizes which LHS models force commuting SEO, or a restriction to nondegenerate full-rank ρ_B with rank-one hidden states. As it stands, it should be rejected, not desk-rejected.","headline":"The central iff between coherence and SDI steering fails: the converse proof assumes decompositions of the identity into ≤d_A positives must be eigenprojectors, which is false.","tokens_in":694,"tokens_out":2107,"would_cite":false,"duration_ms":43469,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P15","81P40","81P45"],"pacs":["03.67.-a","03.65.Ta","03.67.Mn"],"model":"deepseek-v4-flash","headline":"A set of generalized measurements demonstrates semi-device-independent steering if and only if it is coherent, giving coherence a concrete operational meaning in steering tasks.","keywords":["coherent measurements","semi-device-independent steering","measurement incompatibility","quantum steering","quantum randomness","steering-equivalence observables","nonconvex resource theory","detection efficiency"],"falsifier":"Find a state assemblage that admits a dimensionally-restricted LHS model with d_λ ≤ d_A but whose steering-equivalence observables do not commute; for instance, a qutrit LHS model with three hidden states not all diagonal in a common basis would directly contradict Corollary 2.","tokens_in":13882,"feed_emoji":"🎲","tokens_out":4540,"duration_ms":38398,"temperature":0.7,"pith_summary":"This paper aims to give a complete operational characterization of coherent measurements—measurements that do not commute—using semi-device-independent (SDI) steering. The central claim is that a set of generalized measurements can demonstrate SDI steering if and only if it is coherent. This equivalence makes coherence, rather than the stronger property of measurement incompatibility, the fundamental resource for steering under a dimension restriction on the untrusted party. The paper then builds a nonconvex resource theory for SDI steering, constructs a monotone in the two-setting case, and shows that this monotone can certify genuine randomness from states that are separable or unsteerable in the standard one-sided device-independent sense. A striking consequence is that even with arbitrarily low detection efficiency, coherent measurements yield nonzero certified randomness from isotropic states.","feed_headline":"Coherent measurements fully determine semi-device-independent steering","feed_subtitle":"A new equivalence shows coherent measurements certify quantum randomness even from separable states and weak detectors.","key_machinery":"Steering-equivalence observables (SEO): the operators B_{a|x} = ρ_B^{-1/2} σ_{a|x} ρ_B^{-1/2} mapping Bob's unnormalized conditional states back to a measurement-like object. The proof works by showing that a dimensionally-restricted local-hidden-state model exists if and only if these SEO commute; the paper uses this bridge to transfer the property of coherence (pairwise noncommutativity) of Alice's measurement assemblage to the steerability of the state assemblage. The quantification uses the Schatten p-norm of commutators, Υ_p, as the basis for the monotone S_Υ.","core_discovery":"The paper's central discovery is the equivalence: in the one-sided semi-device-independent (1SSDI) scenario, where the dimension of Alice's subsystem is fixed but her measurements are uncharacterized, a state assemblage loses SDI steering precisely when its steering-equivalence observables (SEO) commute. Since an assemblage's SEO inherit noncommutativity exactly from the coherence of Alice's POVMs, the paper concludes that coherent measurements are necessary and sufficient for SDI steering. This gives a one-to-one mapping from any set of generalized measurements to a steering witness, and it shows that the notion of coherence—pairwise noncommutativity—is the operationally relevant property f","pith_inferences":["If the equivalence holds, then in the semi-device-independent setting, any operation that cannot create coherence should be a free operation, suggesting that the resource theory of measurement coherence and SDI steering are essentially the same.","The detection-efficiency result implies that SDI QRNG protocols may be robust against detector inefficiency without active countermeasures, which could be tested in photonic experiments.","The argument extends the domain of quantum randomness to states with quantum discord, so it may connect to other discord-based tasks and suggest new protocols for randomness expansion from noisy states.","A natural next step would be to explore whether the monotone S_Υ can be generalized to more than two settings, or to multi-partite steering scenarios."],"forward_implications":["Any coherent measurement assemblage can be turned into a demonstration of SDI steering by choosing an appropriate entangled state.","SDI steering can be certified directly from the noncommutativity of the SEO, giving a tight criterion for the resource.","A nonconvex resource theory for SDI steering exists, with a faithful, monotonic, nonconvex measure in the two-setting case.","Randomness can be certified and quantified from two-qubit isotropic states even when they are separable or standardly unsteerable, and for any nonzero detection efficiency.","The randomness inequality p_g ≤ 1/2(1+sqrt(1-S_Υ^2)) provides a quantitative bound on the guessing probability in terms of the steering monotone."],"fun_headline_variants":["Coherent measurements are necessary and sufficient for SDI steering","Randomness from coherent measurements, no entanglement required","Coherence fully characterizes semi-device-independent steering","Any coherent measurement set enables steering-based randomness","SDI steering is completely determined by measurement coherence"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The proof's converse direction assumes that every dimensionally-restricted local-hidden-state model has hidden states of the specific form ρ_B^{1/2}|λ><λ|ρ_B^{1/2} in a common basis diagonalizing ρ_B, a premise that may fail for arbitrary decompositions of the identity.","fun_headline_variants_meta":{"raw":{"variants":["Coherent measurements are necessary and sufficient for SDI steering","Randomness from coherent measurements, no entanglement required","Coherence fully characterizes semi-device-independent steering","Any coherent measurement set enables steering-based randomness","SDI steering is completely determined by measurement coherence"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001249,"raw_usage":{"total_tokens":4952,"prompt_tokens":730,"completion_tokens":4222,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":474,"completion_tokens_details":{"reasoning_tokens":4150}},"tokens_in":474,"tokens_out":4222,"duration_ms":29334,"temperature":1.0,"reasoning_tokens":4150,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T22:42:08.121382+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find a state assemblage that admits a dimensionally-restricted LHS model with d_λ ≤ d_A but whose steering-equivalence observables do not commute; for instance, a qutrit LHS model with three hidden states not all diagonal in a common basis would directly contradict Corollary 2.","supporting_citations":[],"review_version":1}